Supplementary Material: Modeling the impact of social distancing,
testing, contact tracing and household quarantine on second-wave
scenarios of the COVID-19 epidemic
Alberto Aleta, David Mart́ın-Corral, Ana Pastore y Piontti, Marco Ajelli, Maria Litvinova,Matteo Chinazzi, Natalie E. Dean, M. Elizabeth Halloran, Ira M. Longini, Jr., Stefano Merler,
Alex Pentland, Alessandro Vespignani, Esteban Moro & Yamir Moreno
May 7, 2020
Contents
1 Mobility data 2
2 Network structure 2
2.1 Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Contacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Social distancing policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.4 Network calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 COVID-19 model 4
3.1 Estimation of the effective reproduction number . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Strategy implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Distribution of infections 8
5 Household attack rate 9
6 Sensitivity analysis 9
6.1 Model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
6.2 Symptomatic detection rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
6.3 Effect of different attack rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
7 School reopening 10
2 NETWORK STRUCTURE
1 Mobility data
The mobility data was obtained from Cuebiq, a location intelligence and measurement company. Thedataset consists of anonymized records of GPS locations from users that opted-in to share the data anony-mously in the Boston metropolitan area over a period of 6 months, from October 2016 to March 2017.Data was shared in 2017 under a strict contract with Cuebiq through their Data for Good program wherethey provide access to de-identified and privacy-enhanced mobility data for academic research and human-itarian initiatives only. All researchers were contractually obligated to not share data further or to attemptto de-identify data. Mobility data is derived from users who opted in to share their data anonymouslythrough a General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA)compliant framework.
From the data we extracted the ”stays”, as the places where anonymous users stayed (stopped) for atleast 5 minutes. Some of the stays happen within places (Points of Interest). We use a dataset of 86kPoints of Interest in the Boston metropolitan area collected using the Foursquare API. Stays are thenaggregated at place level. Finally we estimate the home Census Block Group of the anonymous users asthat in which they are more likely located during nighttime. This results in a dataset of the places peoplestayed including the points of interest that anonymous users visited and the most likely census block groupof where the device owner lives.
2 Network structure
2.1 Agents
Our synthetic population consists of 85k agents (2% of the population in the Boston Metropolitan Area),64k (75%) of them are adults and 21k (25%) corresponds to children. Adults are sampled from the devices inthe mobility data and children are build synthetically from tables B01001 and B11003 from the US Censusdata[?]. All of them together form 43,167 households distributed as follows: 23,293 (53.9%) householdswith only one agent, 7,886 (18.2%) with two agents, 4,959 (11.4%) with three agents, 4,486 (10.4%) withfour agents, 1,784 (4.1%) with five agents, 514 (1.2%) with six agents, and finally, 245 (0.5%) with sevenagents.
Our sample from devices is very representative of the population in the Boston area. As we can see inFigure 1, both population and number of anonymous devices detected in the real data by census area arehighly correlated: ρ = 0.8 (Pearson correlation) with a CI between 0.77 and 0.82 for county subdivisions.
0.01
0.03
0.10
0.30
0.01 0.03 0.10 0.30Percentage of Population by County Subdivision
Perc
enta
ge o
f dev
ices
by C
ount
y S
ubdi
visi
on
ρ=0.80 [0.77,0.82]
Figure 1: The correlation between the population for each county subdivision and the number of devicesin our dataset.
2
2 NETWORK STRUCTURE
2.2 Contacts
In the unmitigated scenario, our network has a total number of 5,029,888 unique daily contacts, 3,924,694(78%) of them in the community layer obtained using the mobility data, and we build synthetically 160,748(3%) for the households layer and 944,446 (19%) for the schools layer. The community layer is based onestimation of co-presence of two devices in Points of Interest visited by the anonymous users (see Methodsin the main paper). Points of interest (POIs) are categorized using the Foursquare taxonomy of placeswhich has ten main categories: Art & Entertainment (4.4%), Colleges & Universities (4.8%), Food (16.7%),Nightlife Spots (3.9%), Outdoors & Recreation (10.6%), Professional & Other Places (23.7%), Shops &Services (29.1%) and Travel & Transport (6.4%), and 638 subcategories. See [?] for a complete list of them.
2.3 Social distancing policies
We simulated two different scenarios for social distancing policies. This produces three contact networks:i) baseline, ii) medium closure, and iii) non-essential closure, as we see it in more detail in Table 1.Schools are closed in the medium and non-essential closure, but both policies differ in the number ofplaces kept open in the community layer. In Table 2 we can see the distribution of POIs, by mainFoursquare category, that remain open during each social distancing policy. In the baseline scenario, wekeep all the categories and thus the average number of contacts in the community layer is 63 (median47, [15-150] 90% confidence interval), with few anonymous individuals having a large number of contacts(that could eventually lead to super-spreading events). In the medium closure scenario, POIs in the Art &Entertainment, Restaurants and Nightlife categories are closed; this drastically reduces the average numberof contacts to 27 (median 15, [0-92] 90%CI). Lastly, when all non-essential places are closed, we only keepopen the following subcategories: Hospital, Salon / Barbershop, Grocery Store, Dispensary, Supermarket,Pet Store, Pharmacy, Urgent Care Center, Dry Cleaner, Drugstore, Maternity Clinic, Medical Supply, andGas Station. In this situation, the average number of contacts is reduced to 6 (median 0, [0-29] 90%CI).The distribution for the number of contacts in the community layer in these three scenarios is shown inFigure 2.
Baseline Medium closure Non-essential closureLayers Contacts %. Contacts % % Diff. Contacts % % Diff.
Community 3,924,694 78 1,378,054 27.4 -72.6 357,144 7.1 -92.9Households 160,748 3.2 160,748 3.2 0 160,748 3.2 0Schools 944,446 18.8 0 0 -100 0 0 -100
Total 5,029,888 100 1,538,802 30.6 -69.4 517,892 10.3 -89.7
Table 1: Number of daily contacts by layer and social distancing policy.
2.4 Network calibration
In general, the probability that i transmits the disease to j can be expressed as
βij = λwij , (1)
where λ is the transmission probability per contact and wij the weight associated to the link between iand j. Hence, we can interpret this expression as a two step-process, first, with probability wij the contactbetween i and j will actually take place and, second, with probability λ the infection will be transmitted.As such, even though the average number of possible contacts (links) that each node has in the communityis 63, the actual average of daily effective contacts under this assumption is 2.50. Similarly, by construction,
3
3 COVID-19 MODEL
Baseline Medium closure. Non-essential closure.POIs categories Open %. Open % % Diff. Open % % Diff.
Arts and Entertainment 3,692 4.44 0 0 -100 0 0 -100Colleges and Universities 4,016 4.83 4,016 4.83 0 0 0 -100Restaurants 13,860 16.7 0 0 -100 0 0 -100Nightlife Spots 3,288 3.95 0 0 -100 0 0 -100Outdoors and Recreation 8,840 10.64 8,840 10.64 0 229 0.27 -97.4Professional and Other Places 19,692 23.71 19,692 23.71 0 415 0.5 -97.8Shops and Services 24,310 29.27 24,310 29.27 0 5,139 6.19 -78.8Travel and Transport 5,370 6.46 5,370 6.46 0 0 0 -100
Total 83,608 100 62,228 74.91 -25.6 5,783 6.96 -93.1
Table 2: Number of POIs open in the Community layer by the different social distancing measures and fullnon-essential closure. Percentages are calculated with respect to the total number of POIs in the baseline.
0.00
0.04
0.08
0.12
0.16
0 50 100 150 200 250 300+
Contacts
Freq
uenc
y
Unmitigated
0.0
0.1
0.2
0 50 100 150 200 250 300+
Contacts
Freq
uenc
y
Medium closure
0.0
0.2
0.4
0.6
0.8
0 50 100 150 200 250 300+
Contacts
Freq
uenc
y
Non−essential closure
Figure 2: Degree distribution in the community layer under normal conditions, soft social distancingmeasures and full non-essential closure.
this number will be 1 both in the household and school layers since we have set the weight of links in theselayers to be 1/(n− 1).
Based on the analysis of contact survey data from 9 countries, the estimated number of daily contactsin Boston is 10.86 in the community+workplace layer, 4.11 in the household layer and 11.41 in the schoollayer [?, ?, ?, ?, ?]. To introduce this into the model, we associate to each layer a single weight, wl, thatwill serve to rescale the weights of each link so that the average number of effective contacts matches thisestimation. As such, we set wC+W = 10.86/2.50, wH = 4.11 and wS = 11.41. These weights multiply theweight of each link depending on the layer they are set in.
3 COVID-19 model
The values of all the disease parameters used for simulating the transmission dynamics are given in table 3.Figure 3 shows the numerical distributions of these parameters as resulting from simulations of the model.
4
3 COVID-19 MODEL
Parameters Description Age group Value Ref.
r relative infectiousness of asymptomatic individuals - 50% †
ε−1 latent period - 3 days [?]
ε′−1 latent period - 5 days [?]
p proportion of asymptomatic - 25% [?]
γ−1 pre-symptomatic period - 2 days [?]
µ−1 time to removed/home stay - 2.5 days ∗
α symptomatic case hospitalization ratio (%) 0-4 0.0 [?]5-17 0.02518-49 2.67250-64 9.33465+ 15.465
χ ICU % among hospitalized 0-4 5.0 [?]5-17 5.018-49 5.3850-64 17.1065+ 44.71
δ−1 days from home stay to hospital admission - 2 [?]
µ−1H days in hospital - 8 [?]
µ−1ICU days in ICU - 13 [?]
k proportion of presymptomatic transmission - 15% [?]
R0 basic reproduction number - 2.5 †
β transmission for symptomatic and asymptomatic individuals - R0µpr+(1−p)/(1−k)
βS transmission for pre-symptomatic individuals - βγkµ(1−k)
Table 3: Baseline set of parameters. †: assumed (see Section 6 for sensitivity analysis);∗: calibrated to thegeneration time Tg.
5
3 COVID-19 MODEL
0
25
50
75
100
A S
Frac
tion
(%
)
Proportion of asymptomatic (p)
0
25
50
75
100
0 5 10 15Days
Frac
tion
(%
)
Latent period (ε)
0
25
50
75
100
0.0 2.5 5.0 7.5 10.0Days
Frac
tion
(%
)
Pre−symptomatic period (γ)
0
25
50
75
100
0.0 2.5 5.0 7.5 10.0Days
Frac
tion
(%
)
Asymptomatic time to recover (µ)
0
25
50
75
100
0.0 2.5 5.0 7.5 10.0Days
Frac
tion
(%
)
Sympt. time to recover (µ) no H.
0
25
50
75
100
0.0 2.5 5.0 7.5 10.0Days
Frac
tion
(%
)Time from Home to Hosp. ( δ)
0
5
10
15
20
25
0−4 5−17 18−49 50−64 65+Age
Frac
tion
(%
)
Hospitalization ( α)
0
10
20
30
40
0−4 5−17 18−49 50−64 65+Age
Frac
tion
(%
)
ICU (%) ( χ)
0.0
2.5
5.0
7.5
10.0
12.5
0 10 20 30 40 50Days
Tota
l ind
ivid
uals 5−17
18−4950−6465+
Time to recover in hospital (µH)
0
2
4
6
0 10 20 30 40 50Days
Tota
l ind
ivid
uals 5−17
18−4950−6465+
Sympt. time to recover (δ+µH) H.
0
2
4
6
0 10 20 30 40 50Days
Tota
l ind
ivid
uals 5−17
18−4950−6465+
Time to recover in ICU ( µICU)
0
1
2
3
4
5
0 10 20 30 40 50Days
Tota
l ind
ivid
uals 5−17
18−4950−6465+
Sympt. time to recover (δ+µICU) ICU
0
5000
10000
15000
20000
0 5 10 15
Frac
tion
(%
)
R0 index (measured)
0
25
50
75
100
A S
Frac
tion
(%
)
Proportion of infections (pt)
0
25
50
75
100
S Pre−S
Frac
tion
(%
)
Proportion of pre−symptomatic infections (k)
0.0
2.5
5.0
7.5
3.0 3.5 4.0 4.5 5.0Days
Frac
tion
(%
)
Td (measured)
Figure 3: Model’s parameters and distributions summary.
3.1 Estimation of the effective reproduction number
We assume that the daily number of new infectious individuals C(t) at time t can be approximated as
C(t) ≈ Pois
(R(t)
t∑s=1
φ(s)C(t− s)
), (2)
where φ is the generation time distribution and R(t) is the effective reproduction number at time t. Thegeneration time is directly measured in the simulations and fitted to a log-normal distribution with mean6.6 days.
In each single run, the likelihood L of the observed time series of cases from day 1 to T is given by
L =T∏t=1
P
(C(t), R(t)
t∑s=1
φ(s)C(t− s)
), (3)
where P (k, λ) is the probability mass function of a Poisson distribution. The posterior distribution of R(t)is then explored using MCMC sampling [?, ?].
After the implementation of social distancing policies, a certain amount of nodes will be completelydisconnected from the system (individuals living alone who lose all their links in the community layer).Denoting this quantity as Z(t), the actual number of infectious individuals that can produce new cases is
6
3 COVID-19 MODEL
C(t)− Z(t). We modify expression (3) accordingly so that
L =T∏t=1
P
(C(t), R(t)
t∑s=1
φ(s) [C(t− s)− Z(t− s)]
). (4)
Note that Z(t − s) will always be 0 in the unmitigated scenario and larger than zero only in a few stepsafter the total closure.
In Figure 4 we compare the effective reproduction number estimated using equations (3) and (4). Rightafter the closure the estimation obtained using (3) shows a small valley followed by an increment of R(t).This is produced by nodes that lose all their connections and thus cannot spread the disease at all. Oncethey recover, since the disease will be still spreading through a small fraction of the system, R(t) seems toincrease. Using equation (4) corrects this problem and R(t) shows a continuously decreasing trend untilthe partial reopening.
Non−essential closure
Partial reopening
Total reopening
0
2
4
6
8
10
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t)
95% C.I.
Mean value
a
Non−essential closure
Partial reopening
Total reopening
0
2
4
6
8
10
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t)
95% C.I.
Mean value
b
Figure 4: Estimation of the effective reproduction number using equation (3) (panel a) or (4) (panel b).Solid lines represent the mean and the shaded region the 95% C.I.
3.2 Strategy implementation
On March 17th all links in the school layer are removed, as well as all links in the community+workplacelayer associated to non-essential places. After 8 weeks, all these links in the community+workplace layerare restored, except the ones belonging to restaurants, nightlife, museums, etc (see Section 2.3 above).Lastly, after 4 weeks the rest of the links of the community+workplace layer are added to the system(while schools remain closed). We align the computational step with the real dates so that on March 17ththe attack rate in the population is 1.5% in the Figures shown in the main text. In section 6.3 we explorethe effect of choosing different attack rates for initializing our simulations.
Then, in the LET scenario, quarantines are applied starting on the same day of the partial reopening(i.e., 8 weeks after the initial closure). We have explored 3 types of quarantines:
i) A fraction qH of all the symptomatic individuals is identified after, on average, qI days of the onset ofsymptoms. After identification, they are completely isolated in special locations (i.e., outside of theirhouseholds). A fraction qT of their contacts are traced and quarantined without testing in speciallocations where they are also completely isolated.
ii) A fraction qH of all the symptomatic individuals is identified after, on average, qI days of the onsetof symptoms. After identification, they are isolated at their homes. A fraction qT of their contactsare traced and quarantined without testing at their homes.
7
4 DISTRIBUTION OF INFECTIONS
iii) A fraction qH of all the symptomatic individuals is identified after, on average, qI days of the onsetof symptoms. After identification, they are isolated in their households and the rest of the householdis set in quarantine (i.e., they cannot contact anyone outside their household). A fraction qT of theircontacts are traced and their households are quarantined without testing.
After 14 days, all individuals isolated or under quarantine are allowed to interact again, regardless of theirhealth status.
In Figure 5, we explore the ICU usage at the peak as a function of the amount of detection (qH) andthe amount of tracing (qT ) when symptomatic individuals are identified, on average, in 1, 2 or 3 days (qI).The dashed line represents the boundary of ICU saturation. For any pair of qT, qH values on the left of thedashed line, the burden of ICU usage will surpass the current capacity. Note that the maximum capacityis the same in all cases, but different values of detection and tracing produce a different maximum usage.These results indicate that investing in contact tracing and symptomatic detection is a better strategythan simply multiplying the number of available ICU beds. Indeed, doubling the ICU capacity would onlyreduce the amount of symptomatic detection necessary by 10%.
We further explore the value of this threshold in Figure 6. We observe that increasing the time ofsymptomatic detection and isolation increases both the number of symptomatic individuals that have tobe detected and the amount of tracing performed. We also note that even though the quarantine of wholehouseholds seems to be the best option, isolating and quarantining individuals outside of their householdsis also a viable strategy. However, as we show next, even though this measure has a lower burden on thefamilies and their individuals economy, it will increase the costs for the overall society.
In Figure 7, we show the maximum number of individuals whose contacts are traced in a single day.Furthermore, note that identifying those individuals as symptomatic also requires testing. Hence, thismeasure will be proportional to the amount of tests that have to be performed daily (the total numberwill depend on the positive test rate). Note that even though isolating and quarantining individuals inspecial locations has a similar effect as doing so in their households from the ICU usage point of view, thenumber of tracers needed in the former strategy is larger than in the latter. This implies that the amount oftesting will also be higher. Added to the cost of creating and managing such special facilities, these resultsindicate that quarantining whole households is the best strategy overall. However, whenever possible, thepossibility of isolating and quarantining single individuals in special locations should be offered to thefamilies who would suffer the most due to the complete quarantining of their household members.
4 Distribution of infections
In Figure 8 we show how the infections that are produced for each strategy are distributed across layers.In the unmitigated scenario the disease spreads initially through the community+workplace layer. By thetime the incidence peaks (around 04/12) infections are transmitted roughly equally through each layer.
In the LIFT scenario the fraction of infections that take place in households is larger than in theunmitigated one during the first closure. Once places in the community+workplace layer are reopened, thedisease starts to spread through that layer again.
Lastly, in the LET strategy, in comparison to the LIFT scenario, we observe a larger fraction ofinfections within households in the period between partial and total reopening, which is to be expectedsince symptomatic individuals are force to stay at home. This suggests that providing special facilities toisolate those individuals will be beneficial for their families.
8
6 SENSITIVITY ANALYSIS
5 Household attack rate
In Figure 9, we explore the secondary attack rate in households. The attack rate in a household, with atleast one infected individual, is defined as the fraction of household members (excluding the first infectedmember) that have suffered the disease. While in the unmitigated scenario this attack rate quickly risesto 80%, in the two other scenarios the situation is slightly different. The initial confinement of individualsat their household initially increases the attack rate, but since the overall burden of the disease is lowerby the end of the epidemic, the final attack rate is also lower than in the unmitigated scenario. This alsoexplains why the household attack rate in the LET scenario is small (compared to the other two situations)even though people are forced to stay at home.
6 Sensitivity analysis
6.1 Model parameters
We conducted a sensitive analysis to study the effects of different percentages of asymptomatic cases andhigher transmissibility. We consider a scenario where all parameters are kept the same as the unmitigatedscenario discussed in the main manuscript with the exception that we consider 50% of cases are asymp-tomatic (henceforth, set A). We also considered a third scenario of higher transmission, where R0 = 3.0and the time to hospitalization is increased δ−1 to 4 days (henceforth, set B). Figure 10 shows the resultsfor the sensitivity analysis when we consider the unmitigated scenario (a-c) and the LIFT scenario (d-e).In both cases the fraction of asymptomatic individuals does not affect the number of new total infections.The change in R0, as expected, does yield a higher number of total infections.
Figure 11 shows the results for the LET scenario where we consider 50% detection of symptomaticcases and different levels of contact tracing: no tracing, 20%, and 40%. The increase of asymptomaticindividuals (set A) reduces the effect of contact tracing. This is even more noticeable when we increaseR0. Although the final numbers differ according to the level of contact tracing, fraction of asymptomatic,and transmission, the same general behavior is observed across scenarios, showing the robustness of ourresults.
Figure 12 shows the evolution of hospitalizations for the different scenarios. When we compare thenew results from the sensitive analysis with the ones reported in the main manuscript, we observe that thetiming of the peak on the ICU needs is the same, while the number of ICU needed is higher in a highertransmission situation.
Summarizing, the sensitivity analysis shows that the modeling results presented in the main text arerobust to plausible ranges of parameter values for the key time-to-event intervals of COVID-19.
6.2 Symptomatic detection rate
In the main text we mainly explored the situation in which 50% of all symptomatic individuals can bedetected and their contacts traced, on average, in two days. In Figures 13 and 14 we show the evolutionof the system with a detection rate of 30% and 70%, respectively. Once again, our results and conclusionshold.
6.3 Effect of different attack rates
In Figure 15 we explore the effect of choosing a different initial attack rate for the implementation of thesocial distancing policies. We observe that the sooner these policies are implemented (i.e., the disease has
9
7 SCHOOL REOPENING
not spread through the population and thus the attack rate is low) the smaller the first peak of the diseasewill be. However, once the restrictions are removed, if there is a second wave the peak will be larger.Conversely, if the policies are implemented once a large fraction of the population has already suffered thedisease, the second wave of the epidemic will be much milder.
7 School reopening
In the main scenarios studied in the paper, once schools are closed they remain so indefinitely. In thissection, we explore the consequences of reopening schools on the 15th of September. In figure 16 we showthe evolution of the epidemic for the LIFT and LET scenarios under such condition. As we can see, sincein the LIFT scenario the peak of the epidemic occurs in July, the effect of reopening schools is negligible.A similar observation can be made in the LET scenario without tracing. However, for the LET scenarioswith tracing, since the peak of the epidemic has not faded out by that date, reopening schools producesa small increment on the incidence. Nonetheless, in figure 17 we observe that such increment does notsaturate the health care system. This signals that the increase in the number of infections is due to childrengetting infected, since this group is less likely to require hospitalization.
In figure 18 we explore the affordability of the LET scenario when schools are reopened. In the cases ofno tracing and 20% tracing, since the peak of quarantined households is set before September, the reopeningdoes not produce a large increase on the households that have to be set under quarantine. However, inthe 40% tracing scenario since the peak is close to the date when schools would reopen, the populationunder quarantine could pass the 10%. It is important to stress that the role of children in the transmissionof SARS-Cov-2 is not yet clear. Several papers indicates a differential susceptibility to the infection aswell as a smaller probability of developing symptoms. In principle this might lead smaller role of schoolreopenings that should be assessed in the presence of firm data on differential forward transmission acrossdifferent age brackets.
Lastly, in figure 19 we show the household attack rate and the fraction of infections that would takeplace in each layer. We observe that the household attack rate would remain stable and that the numberof infections that would take place in schools will increase above 0.
10
7 SCHOOL REOPENING
Isolate/Quarantine individuals outside households
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 1 daya
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 2 daysb
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 3 daysc
Isolate/Quarantine individuals in households
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 1 dayd
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 2 dayse
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 3 daysf
Isolate/Quarantine whole households
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 1 dayg
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 2 daysh
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 3 daysi
Half current capacity Current capacity Twice current capacity Ten times current capacity
ICU usage at the peak
Figure 5: Maximum ICU occupation with different isolation/quarantine strategies. From left to right, eachcolumn shows the results when the symptomatic individuals are detected and isolated after 1, 2 or 3 days.From top to bottom, each row shows the results for strategies i), ii) and iii). The dashed line indicateswhen the ICU occupation excess the current availability.
11
7 SCHOOL REOPENING
ICUsaturation
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 1 daya
ICUsaturation
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 2 daysb
ICUsaturation
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 3 daysc
Inside households Outside households Whole households
Figure 6: Threshold for ICU saturation - with the current estimated ICU availability - as a function of thefraction of symptomatic individuals detected and the fraction of contacts traced. From left to right, weshow the results when the symptomatic individuals are detected and isolated after 1, 2 or 3 days.
12
7 SCHOOL REOPENING
Isolate/Quarantine individuals outside households
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 1 daya
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 2 daysb
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 3 daysc
Isolate/Quarantine individuals in households
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 1 dayd
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 2 dayse
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 3 daysf
Isolate/Quarantine whole households
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 1 dayg
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 2 daysh
0
20
40
60
80
100
0 20 40 60 80 100Symptomatics detected (%)
Con
tact
s tr
aced
(%
)
Detection in 3 daysi
0.06 0.20 0.50 1.00 2.00 4.00 8.00
Maximum daily tracings per 1,000 people
Figure 7: Maximum number of individuals whose contacts are traced in a single day per 1,000 people.From left to right, each column shows the results when the symptomatic individuals are detected andisolated after 1, 2 or 3 days. From top to bottom, each row shows the results for strategies i), ii) and iii).
13
7 SCHOOL REOPENING
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/02
Frac
tion
of
infe
ctio
ns (
%)
C+WHS
UnmitigatedaNon−essential closure
Partial reopeningTotal reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/27
C+WHS
LIFTbNon−essential closure
Partial reopening + Q.Total reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/22
C+WHS
LET (Detection 50% + Tracing 40%)c
Frac
tion
of
infe
ctio
ns (
%)
Frac
tion
of
infe
ctio
ns (
%)
Figure 8: Distribution of infections across layers in the unmitigated, LIFT and LET (with 50% detectionof symptomatic individuals and 40% tracing) scenarios. Lines represent the average fraction of infectionsthat take place in each layer out of all the infections that have been produced each day.
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/02
Hou
seho
ld a
ttac
k ra
te (
%)
aNon−essential closure
Partial reopeningTotal reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/27
Hou
seho
ld a
ttac
k ra
te (
%)
bNon−essential closure
Partial reopening + Q.Total reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/22
Hou
seho
ld a
ttac
k ra
te (
%)
cUnmitigated LIFT LET (Detection 50% + Tracing 40%)
Figure 9: Household attack rate as a function of time in the unmitigated scenario, a, LIFT strategy, b,and LET strategy with 50% detection of symptomatic individuals and 40% tracing, c. Solid lines representthe average value while the shaded region shows the 95% C.I.
0
10
20
30
40
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
Default setSet ASet B
New infectionsa
0
250
500
750
1000
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Cum
ulat
ive
infe
ctio
ns/1
,00
0 p
eopl
e
Default setSet ASet B
Total infectionsb
0
1
2
3
4
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
R(t
)
Default setSet ASet B
E�ective reproduction numberc
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng
Tota
l reo
peni
ng
0
3
6
9
12
15
18
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
Default setSet ASet B
New infectionsd
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ngTo
tal r
eope
ning
0
250
500
750
1000
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Cum
ulat
ive
infe
ctio
ns/1
,00
0 p
eopl
e
Default setSet ASet B
Total infectionseNon−essential closure
Partial reopeningTotal reopening
0
1
2
3
4
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/2001/
17
R(t
)
Default setSet ASet B
E�ective reproduction numberf
Figure 10: Evolution under different sets of parameters for the unmitigated scenario and LIFT strategy.
14
7 SCHOOL REOPENING
No
Trac
ing
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0
2
4
6
8
10
12
14
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
a
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng0
250
500
750
1000
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Cum
ulat
ive
infe
ctio
ns/1
,00
0 p
eopl
e
bNon−essential closure
Partial reopening + Q.
Total reopening
0
1
2
3
4
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t
)
c
20%
Tra
cing
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0.0
1.5
3.0
4.5
6.0
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
01/17
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
d
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0
250
500
750
1000
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Cum
ulat
ive
infe
ctio
ns/1
,00
0 p
eopl
e
eNon−essential closure
Partial reopening + Q.
Total reopening
0
1
2
3
4
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t
)
f
40
% T
raci
ng
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0.0
1.5
3.0
4.5
6.0
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
g
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0
250
500
750
1000
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Cum
ulat
ive
infe
ctio
ns/1
,00
0 p
eopl
e
hNon−essential closure
Partial reopening + Q.
Total reopening
0
1
2
3
4
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t
)
i
Default set Set A Set B
Figure 11: Evolution under different sets of parameters for the LET scenario with 50% of symptomaticdetection. The top, mid and bottom rows show the results with 0%, 20% and 40% of tracing, respectively.
15
7 SCHOOL REOPENING
0
3
6
9
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Unmitigateda
0.0
1.5
3.0
4.5
6.0
7.5
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
LIFTb
0.0
0.3
0.6
0.9
1.2
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
LET 20% tracingc
0.0
0.3
0.6
0.9
1.2
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
LET 40% tracingd
0
3
6
9
12
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Unmitigatede
0.0
1.5
3.0
4.5
6.0
7.5
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
LIFTf
0.0
0.3
0.6
0.9
1.2
1.5
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
LET 20% tracingg
0.0
0.3
0.6
0.9
1.2
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
LET 40% tracingh
Figure 12: Evolution of the hospitalizations. Top row: set of parameters A (50% of asymptomatic indi-viduals). Bottom row: set of parameters B (25% of asymptomatic individuals, R = 3.0, δ−1 = 4 days).
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
02468
101214
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
no tracing20% tracing60% tracing
a
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng+
Q.
Tota
l reo
peni
ng
0
250
500
750
1000
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Cum
ulat
ive
infe
ctio
ns/1
,00
0 p
eopl
e
no tracing20% tracing60% tracing
b
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0
5
10
15
20
02/1604/1206/0708/0209/2711
/2201/
1703/14
Frac
tion
of
hous
ehol
ds(%
)
no tracing20% tracing60% tracing
Quarantined householdsc
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0
50
100
150
200
02/1604/1206/0708/0209/2711
/2201/
1703/14Q
uara
ntin
ed in
d./1
,00
0 p
eopl
e
Unknown Symptomatic
Quarantined individuals (60% trac.)d
Non−essential closurePartial reopening + Q.
Total reopening
0
1
2
3
4
5
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
R(t
) no tracing20% tracing60% tracing
e
0
2
4
6
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Detection 30% No Tracing f
0.0
0.5
1.0
1.5
2.0
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Detection 30% Tracing 20%g
0.0
0.5
1.0
1.5
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Detection 30% Tracing 60%h
Figure 13: Evolution of the system when 30% of the symptomatic individuals are isolated at their homesand their contacts traced and quarantined in their households, default set of parameters.
16
7 SCHOOL REOPENING
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
02468
101214
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
no tracing20% tracing40% tracing
a
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng+
Q.
Tota
l reo
peni
ng0
250
500
750
1000
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
no tracing20% tracing40% tracing
b
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0
5
10
15
20
02/1604/1206/0708/0209/2711
/2201/
1703/14
Frac
tion
of
hous
ehol
ds(%
)
no tracing20% tracing40% tracing
c
Non
−es
sent
ial c
losu
re
Par
tial
reo
peni
ng +
Q.
Tota
l reo
peni
ng
0
50
100
150
200
02/1604/1206/0708/0209/2711
/2201/
1703/14Q
uara
ntin
ed in
d./1
,00
0 p
eopl
e
Unknown Symptomatic
Quarantined individuals (40% trac.)d
Non−essential closurePartial reopening + Q.
Total reopening
0
1
2
3
4
5
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/2001/
17
R(t
) no tracing20% tracing40% tracing
e
0
2
4
6
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Detection 70% No Tracingf
0.0
0.5
1.0
1.5
2.0
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Detection 70% Tracing 20%g
0.0
0.5
1.0
1.5
02/1603/1504/1205/1006/0707/0508/0208/3009/2710
/2511
/2212
/20
Hos
pita
lized
/1,0
00
peo
ple
HICUICU beds
Detection 70% Tracing 40%h
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
Cum
ulat
ive
infe
ctio
ns/1
,00
0 p
eopl
e
Quarantined households
Figure 14: Evolution of the system when 70% of the symptomatic individuals are isolated at their homesand their contacts traced and quarantined in their households, default set of parameters.
Non−essential closurePartial reopening
Total reopening
0
1
2
3
4
5
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t
)
0.01%3%10%20%
E�ective reproduction numberaNon−essential closure
Partial reopeningTotal reopening
0369
12151821
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Dai
ly in
fect
ion
inci
denc
e/1
,00
0 p
eopl
e
0.01%3%10%20%
New casesbNon−essential closure
Partial reopeningTotal reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Hou
seho
ld a
ttac
k ra
te (
%)
0.01%3%10%20%
Secondary attack ratec
Figure 15: Effective reproduction number, daily incidence and household attack rate for different initialattack rates. The sooner the intervention, the larger the second wave can be.
17
7 SCHOOL REOPENING
Non
−es
sent
ial c
losu
re
Par
tial r
eope
ning
Tota
l reo
peni
ng
Sch
ool r
eope
ning
0
3
6
9
12
15
18
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
01/17
Dai
ly in
fect
ion
inci
denc
e/1,
000
peop
le
a
Non
−es
sent
ial c
losu
re
Par
tial r
eope
ning
Tota
l reo
peni
ng
Sch
ool r
eope
ning
0
250
500
750
1000
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Cum
ulat
ive
infe
ctio
n in
cide
nce/
1,00
0 pe
ople
b
Non−essential closure
Partial reopening
Total reopening
School reopening
0
1
2
3
4
5
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t
)
cN
on−
esse
ntia
l clo
sure
Par
tial r
eope
ning
+ Q
.
Tota
l reo
peni
ng
Sch
ool r
eope
ning
0
3
6
9
12
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Dai
ly in
fect
ion
inci
denc
e/1,
000
peop
le
no tracing20% tracing40% tracing
dN
on−
esse
ntia
l clo
sure
Par
tial r
eope
ning
+ Q
.
Tota
l reo
peni
ng
Sch
ool r
eope
ning
0
250
500
750
1000
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Cum
ulat
ive
infe
ctio
n in
cide
nce/
1,00
0 pe
ople no tracing
20% tracing40% tracing
e
Non−essential closure
Partial reopening + Q.
Total reopeningSchool reopening
0
1
2
3
4
5
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
R(t
)
no tracing20% tracing40% tracing
f
Figure 16: Evolution of the LIFT and LET scenarios when schools are reopened on the 15th of September.
0.0
1.5
3.0
4.5
6.0
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Hos
pita
lized
/1,0
00 p
eopl
e
HICUICU beds
LIFTa
0.0
0.2
0.4
0.6
0.8
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Hos
pita
lized
/1,0
00 p
eopl
e HICUICU beds30% ICU beds
LET (Detection 50% + Tracing 20%)b
0.0
0.2
0.4
0.6
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Hos
pita
lized
/1,0
00 p
eopl
e HICUICU beds30% ICU beds
LET (Detection 50% + Tracing 40%)
Figure 17: Evolution of hospitalizations in the LIFT and LET scenarios when schools are reopened on the15th of September.
Non
−es
sent
ial c
losu
re
Par
tial r
eope
ning
+ Q
.
Tota
l reo
peni
ng
Sch
ool r
eope
ning
0
3
6
9
12
02/1604/12
06/0708/02
09/2711/22
01/1703/14
05/09
Fra
ctio
n of
hou
seho
lds
(%) no tracing
20% tracing40% tracing
Quarantined householdsa
Non
−es
sent
ial c
losu
re
Par
tial r
eope
ning
+ Q
.To
tal r
eope
ning
Sch
ool r
eope
ning
0
25
50
75
100
125
02/1604/12
06/0708/02
09/2711/22
01/1703/14
05/09Qua
rant
ined
ind.
/1,0
00 p
eopl
e Unknown Symptomatic
Quarantined individualsb
Non
−es
sent
ial c
losu
re
Par
tial r
eope
ning
+ Q
.To
tal r
eope
ning
Sch
ool r
eope
ning
0.00
0.05
0.10
0.15
0.20
0.25
02/1604/12
06/0708/02
09/2711/22
01/1703/14
05/09
Ind.
trac
ed/1
,000
peo
ple
Daily individuals tracedc
Figure 18: Affordability of the quarantine strategy when schools are reopened on the 15th of September.
18
7 SCHOOL REOPENING
Non−essential closurePartial reopening
Total reopening
School reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/27
Hou
seho
ld a
ttack
rat
e (%
)
LIFTaNon−essential closure
Partial reopening + Q.Total reopening
School reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/22
Hou
seho
ld a
ttack
rat
e (%
)
LET (Detection 50% + Tracing 40%)b
Non−essential closurePartial reopening
Total reopening
School reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Fra
ctio
n of
infe
ctio
ns (
%) C+W
HS
LIFTcNon−essential closure
Partial reopening + Q.Total reopening
School reopening
0
20
40
60
80
100
02/1603/15
04/1205/10
06/0707/05
08/0208/30
09/2710/25
11/2212/20
Fra
ctio
n of
infe
ctio
ns (
%) C+W
HS
LET (Detection 50% + Tracing 40%)d
Figure 19: Household attack rate (a, b) and fraction of infections that take place in each setting (c, d)when schools are reopened on the 15th of September.
19